Formation Evaluation and Behind Casing Opportunity Analysis Using Multi-Output Regression and Machine Learning Techniques

Abstract

Generally, Behind Casing Opportunity (BCO) provides a good alternative to maximize brownfield production by enabling access to the remaining oil targets behind existing completion. Proper BCO maturation which includes minor reservoir unit analysis may offer a high return, low-risk item in terms of obtaining cheap oil at low cost. Unfortunately, a lack of proper resources such as possible manpower and budget constraints may pose a challenge for proper BCO analysis. A novel application of multioutput regression Random Forest algorithm to predict both BCO and fluid code for BCO determination marks a good start for further implementation of machine learning to tackle this problem. The multi-output model enables two or more variables to be predicted hence allowing uniformed prediction rules and time-saving alternatives. Exploratory data analysis (EDA) and required data preprocessing were carried out to provide excellent inputs for the algorithm. The algorithm produced two outputs of predicted BCO and fluid code with root mean squared error (RMSE) of 0.0933 and R2 of 0.9619. To properly support the logic of the model prediction, the log curve of both predicted values was plotted and the rationale behind the predicted result was observed. Besides, a good cross-plot correlation of both the predicted and actual value of the outputs also aided in further validating the result. The research potentially can help to further enhance BCO analysis by giving robust and very effective methods for both BCO and fluid detection. Furthermore, predicting fluid code helps for proper reservoir analysis hence providing a time-saving alternative for better drilling decisions.